一种改进级联U-Net网络的结肠息肉分割算法

结肠镜图像中息肉的精确分割是诊断结肠癌的关键环节，针对目前结肠息肉分割算法存在孔洞、分割粗糙以及分割不完全的问题，提出了一种改进级联U-Net结构的结肠息肉分割算法。运用特征融合思想，设计了多尺度语义嵌入模块和残差模块，充分利用深、浅层特征的语义信息。引入注意力机制，在模型的级联处构建了改进空洞卷积模块，扩大卷积感受野并增强特征捕获能力。改进了卷积层模块和分割损失函数，提升模型的泛化性和鲁棒性。在Kvasir-SEG数据集上进行实验分析，相似系数、平均交并比、召回率和准确率分别达到了90.39%、88.34%、83.62%和95.12%。实验结果表明，该文所提算法改善了分割图像内部孔洞、边缘粗糙及分割不完全的问题，优于其他息肉分割算法。

A colon polyp segmentation algorithm based on improved cascaded U-Net network

Accurate segmentation of polyps in colonoscopy images has become a key aspect in the diagnosis of colon cancer.A colon polyp segmentation algorithm with improved cascade U-Net structure is proposed to address the problems of holes,rough segmentation and incomplete segmentation in the current colon polyp segmentation algorithm.Using the idea of feature fusion,a multi-scale semantic embedding module and a residual module are designed to make full use of the semantic information of deep and shallow features.An attention mechanism is introduced and an improved null convolution module is built at the cascade of the model to expand the convolutional field of perception and enhance feature capture.The convolutional layer module and segmentation loss function are improved to promote the generalization and robustness of the model.The experimental analysis on the Kvasir-SEG dataset achieves 90.39%,88.34%,83.62% and 95.12% for similarity coefficient,average intersection ratio, recall and accuracy, respectively. The experimental results show that the proposed algorithm improves the problems of internal holes,rough edges and incomplete segmentation of segmented images and outperforms other polyp segmentation algorithms.